利用多群心群聚法為基礎之彩色影像分割研究

近年來彩色影像分割已經被廣泛的應用在不同的領域，不再只是應用在影像資料，甚至多媒體的電影、電視、數位相機等。使用群聚法來分割彩色影像是最快速及簡單的方法，而ㄧ般傳統的分群方法並不適合來分割影像，除了難以給予正確的分類數目外，還有影像區域在特徵空間中並非單純圓形分布的問題。因此本論文的主要目的為提出一個適合於彩色特徵空間的群聚演算法，是以模糊聚類(Fuzzy C-means)為量化基礎，利用彩色影像的標準差當做其初始分類數目，再依照彩色影像物體在特徵空間中獨特散佈狀態，將屬於同一群的群心予以合併，一個群的表示法由單群心變為多群心，使得彩色影像不同的區域在特徵空間中能夠較完整分割出來。本演算法具有影像區域將會更完整的分割及不需要正確的指定出群聚的數目等優點。最後使用學者常用的彩色影像加以測試，結果本研究提出的方法比單群心分類方法在分割彩色影像上有更好的效果。

關鍵字

模糊聚類；彩色影像分割；多群心

並列摘要

In recent years, color image segmentation has wide applications in various areas, not only in the area of image data processing, but also in areas of multimedia movie, TV, digital camera, and so on. The clustering method is the one of fastest and simplest methods to segment color images. Traditional clustering methods are difficult to segment images due to the following two reasons: it is hard to define the correct cluster number and image color space distribution is not fixed in a typical shape, such as circle. This motivates us to propose a novel approach for color image segmentation. First, we analyze the RGB color space of an image to find its standard deviation which can be the input of the fuzzy clustering method (Fuzzy C-means) in order to find a rough cluster number. Then this study uses a search rectangle to find the next most suitable center to combine based on the distance between cluster centers. We combine one cluster with another cluster to form multiple centers in one cluster, and the final shape of the cluster can be anything. The advantages of this method are that we can have better color segmentation and we do not need to know the cluster number beforehand. Finally, we use some color images commonly used in the literature to show that the multiple cluster centers method performs better than the single cluster center method for color image segmentation.

並列關鍵字

Multiple cluster centers ； Color image segmentation ； Fuzzy clustering

參考文獻

呂理邦，民國九十二年。利用區域內外判定法則與模糊理論於影像分割之研究，中原大學工業工程研究所碩士，桃園。

Alain Tremeau, Nathalie Borel “A region growing and merging algorithm to color segmentation” Pattern Recognition, vol. 30, pp. 1191-1203, 1997.

Alireza Khotanzad, Orlando J. Hernandez. “color image retrieval using multispectral random field texture model and color content feature” Pattern Recognition, vol. 36, pp.1679-1694, 2003.

Bezdek, J.C. “A convergence theorem for the fuzzy ISODATA clustering algorithms”, IEEE Trans. on Pattern Analysis and Machine Intelligence, vol. 2, pp. 1-8, 1980.

Bezdek, J.C., “Pattern recognition with fuzzy objective function algorithms,” New York: Plenum Press, 1981.

國際替代計量

利用多群心群聚法為基礎之彩色影像分割研究

未授權

主題瀏覽